Effects of Colon Cancer Insurance Policies: Evidence from NHIS

Project Outcomes Statement

After performing a variety of data quality checks to ensure the fidelity of the restricted NHIS data, we were able to complete Specific Aim 1. Specifically, we focused on triple differences models that exploit variation in colon cancer insurance policies at the state by year by age level. Prior research has only been able to study colon cancer insurance mandates using variation at the state by year level due to limitations of existing public use data (namely, that they do not ask colorectal cancer screening questions to individuals below age 50). Because the NHIS asks colorectal cancer (CRC) screening questions to individuals beginning at age 40 despite that most colon cancer insurance policies set eligibility for benefits beginning at age 50, this provides us a within-state control group (40-49 year olds) that allow us a more internally valid way to estimate the causal effects of state level colon cancer insurance policies.

Our final analysis sample restricted attention to NHIS data with information on CRC screenings from around every 3 years from 2000 on. These are the years when the CRC questions were asked and consistent NHIS final weights were available. In addition to considering 40-64 year olds, we also restricted attention to individuals with private health insurance. Since the state level colon cancer insurance mandates we are most interested in only applied to private health plans, any effects – if they exist – have to be driven by privately insured individuals. We note that most existing work has relied on datasets such as the BRFSS which do not identify source or type of insurance for most of the sample period.

We estimated triple difference models of the likelihood of ever having a CRC screening for privately insured 40-64 year olds with detailed controls for individual level demographics (race/ethnicity, sex, education, and marital status), fixed effects for state, year, and 5 year age group, and all two-way combinations of those sets of state/year/age group fixed effects, as well as some continuous state level policy controls. Note that in these models all the characteristics and 0/1 policy indicators that vary only at the state-by-year level fall out of the regression equation because they are perfectly collinear with the state-by-year
fixed effects. In this model with a sample size of N=56,840 and an R-squared of 0.1301, the coefficient estimate on the colon cancer insurance mandate was -0.004 with a standard error of 0.012 (i.e., it was small and not statistically significant). We estimated a range of robustness and specification tests; the mandate estimate continued to be small in magnitude, not statistically significant, and moreover sometimes it was of an unexpected sign.

The findings and conclusions in this report are those of the author(s) and do not necessarily represent the views of the Research Data Center, the National Center for Health Statistics, or the Centers for Disease Control and Prevention.